MPLS traffic engineering

Question

While studying MPLS I couldn't understand it's one basic functionality. Mpls creates tunnel interfaces or pseudowires between label edge routers. To achieve traffic engineering I can set different limits on different tunnels. But what decides which packet takes which tunnel. For instance at the ingress LER there are 3 tunnel interfaces with throughout 50,100,250 mbps respectively. If a packet now enters the mpls network what decides which tunnel it will use .

One option I saw was static which has obvious limitations.

Second option was route maps. Does this mean I can match source IP address and determine tunnel interface ?

Third option is auto and I couldn't understand how it works.

In general does any deep packet inspection technique be used ?

Thanks,

Answer 1

Traffic steering is actually the task that is beyond MPLS TE itself. TE tunnel just provides the path, and it's up to other protocols and techniques to steer actual traffic. The same is true for bandwidth reservation. TE just says "there are 300 mbps reserved", but nothing stops your client from sending more than that.

So, your options are policies, static routes and IGPs. If you want granular control you go with static routes and policies. If you just want to get fast restoration you go with announcing tunnels into IGP. Steering technique depends on your task.

Answer 2

to expand on the previous answer -- policy is generally dictated by qos. an s/p will generally match on dscp or cos bits (cbts or pbts) to decide tunnel path. there are other ways, in which you can match by source/destination addresses, vrf, etc. they usually take the form of some sort of access-list applied as a policy to the tunnel-path. the policy can take the form as a route-map, depending on platform.

q.

Answer 3

If your tunnels follow the same path, but are defined in order to provide several QoS, you should look at Differentiated Service standards and how it integrates with MPLS and Traffic Engineering. Eg. E-LSP where the class of traffic is inferred from the MPLS EXP field, which can be itself inferred from the IP DSCP field at the MPLS network ingress. However in the example, you will have only one tunnel, where traffic classification and priorization is achieved.

If your tunnels follow different path, By default Cisco MPLS routers achieves a weighted load-balancing of the traffic across available path (considering that LSP Tunnels have the same ingress and egress). Weights can be inferred from bandwidth (eg. allocated with RSVP-TE), or administratively configured weights.

As you proposed, you can also use route-maps and ACLs to enforce given flows to be switched through a specific LSP tunnel. However I assume that to stay efficient, packet inspection won't be beyond layer 3 (IP source/destination, IP DSCP/TOS).

Considering your comments, if you need to implement it using BGP, you shoold look at Flowspec. This standards allows distributing filtering rules via BGP. You can configure the action to achieve to the filtered traffic, eg.:

• redirecting to a specific Next Hop,
• redirecting to a given VRF, where wo can configure the chosen Tunnel as Next Hop (by default or not)